FI105248B - A method for transmitting digitized, block-coded audio signals using scaling factors - Google Patents

Abstract

A method of transmitting digitized block coded audio signals includes forming scale factors of the digitized audio signals. The n(k-1) differences are formed from k successively in-time scale factors for each frequency sub-band or for a group of spectral values of the audio signal. The n(k-1) differences are grouped into at least two value classes. New scale factors are selected for each of the n sub-bands or spectral value groups based on a sequence of n(k-1) value classes. Identifying information, including the control information indicating at which locations in the sequence of n(k-1) value classes the selected new scale factors are disposed, is associated with each sequence of n(k-1) value classes. The associated selected new scale factors are assigned to each sequence of the sampled signal values and to the identifying information associated with each sequence of sampled signal values. A transmission pattern of new scale factors is determined separately for each of the n sub-bands or spectral value groups. Lastly, audio signals are regenerated from the sampled signal values and from the assigned selected new scale factors.

Description

1 105248

The invention relates to a method according to the preamble of claim 1. The present invention relates to a method for transmitting digitized, block-coded audio signals using scaling factors. Such a method is known from EP-A-64119.

For block coding of digitized audio signals, it is known (DE-A-33 28 344) to generate a scaling factor from the peak value of a string (block) of signal values, which coefficient indicates the range of the peak value from a plurality of magnitude ranges. As is also known from DE 36 39 753, the sample values of signal 20 may be represented by multiple subband signals of the spectrum, and the sample values of the various subbands may be modified according to the respective hearing threshold of the human hearing to reduce irrelevant and redundant information. Instead of subdividing the spectrum into subbands, the digitized audio signal may also be subjected to Fourier analysis and assigned to the spectral value groups by scaling factors and by performing the reduction of irrelevant and redundant information. Here, a reduction of irrelevant and redundant data for high-quality audio signals, e.g., broadcast signals, is made such that the compressed digital audio signal is suitable for: an ISDN transmission network hierarchy, i.e., having a constant bit rate of n * 64 kbit / s. In order to achieve satisfactory sound quality, especially at low 64 kbit / s bit rates, control information, such as scaling coefficient operations, must be compressed very vigorously with respect to their data rates. EP-A-64,119 is also known in connection with block coding of high band speech signals that 2 105248 is first formed by the difference of a block or subband successive scaling coefficients to reduce the number of scaling coefficients to be transmitted per block or subband. In the event that the predetermined difference range is exceeded or exceeded, all the scaling factors under consideration are transferred, otherwise only the highest of the scaling factors considered. In order to identify on the decoder side how many scaling coefficients applied to the scaling coefficients, the identification information is transmitted in addition to the scaling coefficients.

10

As a result, the known compression of the scaling factors only reduces redundant information, whereas non-essential information is not transmitted based on psycho-acoustic considerations that take into account the pre- and post-suppressor effects of human auditory perception.

The object of the invention, on the other hand, is to perform a data reduction on the scaling factors, whereby in addition to the redundant information, irrelevant information is also reduced.

This object is solved according to the invention by the features of claim 1.

The invention will now be explained in more detail by way of an exemplary embodiment. . 25.

The bit-saving scaling factor transfer is based on pattern recognition and coding, eliminating both redundancy and irrelevance.

30: The input matrix consists of k columns and n rows, where k, represents the number of successive scaling factors and the number of n subbands. From now on, the explanation applies * generally using the example in the left-hand column and 35 in the right-hand column. For the example (right), there are three consecutive scaling factors and a total of 32 subbands (denoted as 3 105248 frames below). The scaling factors are quantized with six bits and can thus have 2 6 = 64 values. The block length for which the scaling factor is generated is 8 ms.

5 Input matrix: scfll scfl2 ..... SCflkT j SCfll SCf 12 SCfl3 “I 3 * 6 bit-18 bit SCf 21 j scf21 I 3 * 6 bit-18 bit ^ scfnl scfn2 ..... scfnkj | [scf31 1 SCfi2 1 SCfi3 3 * 6 bit-18 bit all frames 32 * 18 bit = 578 bit

Next, calculate the difference between the consecutive values of each row and / or column d. They form an intermediate matrix. 15 The size of the differences depends on the absolute length (row) of the block, or the width of the subbands (column). In the following, we consider only the sequential scaling factors over time.

Example: 20 dl, 21 = SCf 12 - scfll dto.

The intermediate matrix has k-1 columns (when looking at the columns there are also n-1 rows).

- 25 ...

• Outdoor matrix: dl, 21 dl, 32 ..... dl, mm Ί I ["dl, 21 dl, 32 d2,21 I | I d2,21

30 j ^ dn'21 .......... d „, kk-l {I JjJ32,21 d32,32 J

• Possible differences Di are divided into classes using Table 1.

Category 1 is a set with one or more differences D.

For the example on the right, the differences fall into five classes of 1, where one or 35 more differences fall into the same class: 4 105248 11 = {Di, D2, De} for d <-2 and d> 64 li 12 * = {D7, De, ... »D12} when d <-3 and d <O I2 13 - {D13, D14,. ·., D17} when d = O I3 1« = {D4} when d <3 and d> O I4 5 when d> 2 and d <64 I5

In "{Di-4, Di_3, ...» Di}

All possible sequences of classes 1 give 1 to the combination c (k-1): 10 ci = li li ... li C2 m li li · · I2 C3 = li li ... l3 C (ic-i) l = lk -1 pp-1 ... ljc-i 15

The data reduction is achieved by combining the scaling factor transfer model with the combinations c in each case. This model consists of control data and a queue of scaling factors, whereby the number of different scaling factors in the queue is less than or 2q equal to the number of different scaling factors in the input matrix. The determination of the transmission pattern, as evidenced by categorized differences, is based on statistical knowledge of the signal and psychoacoustic considerations, and in particular on the pre- and post-attenuation effects of human auditory perception. If, for example, the scaling factor of the input matrix row does not. 25 ί does not change at all, so not all scaling factors need to be transferred because this is redundant information (redundancy). Increasing scaling factors need to be shifted more accurately than decreasing, since it is known from psychoacoustics that the human 3Q auditory state is a clear post-attenuation within an area up to 200 ms, but on the other hand a pre-attenuation within an i <even 20 ms (irrelevance).

An income matrix with, for example, three columns gives 25 line-by-line subtractions two categorical distinctions and 25 composite options c. A scaling factor model is associated with each combination c. This provides a specified number of 105248 5 movable scaling factors and control information about where the scaling factors are located or where they change. In this example, one, two, or three scaling factors and 2 bits of control data must be transferred.

5 class control combination combination transfer model scf-number. information li li ci scfi scf2 scf3 3 (18 bit) 11 (2 bit) li I2 C2 scfi scf2 scf3 2 (12 bit) 01 (2 bit) li I3 C3 scfi scf2 scf3 10 li 14 C4 scfi scf2 scfs ...

11 15 ---- 12 li C6 scfi scf2 scf3 2 (6 bit) 10 (2 bit) I2 lz C7 scfi scfa scf3 1 (6 bit) 00 (2 bit) I2 I3, I2 1 «15 I2 15 la li

IS I5 C2S

This example gives four possible colonies, so the 20 control information is 2 bits: 1. three different scaling factors 81 s2 83 00 25 * «! The amount of data per frame is: (3 * 6 bits + 2 bits) * 32 = 640 bits 2. first scaling factor for first position, 30 second for second and third position • ·

Sl S3 s3 01 sl s2 s2

For each frame, the amount of data is: (2 * 6 bit + 2 bit) * 32 = 448 35 bit 6 105248 3. first scaling factor for first and second position, third for third position s2 s2 s3 10 g sl sl s3

For one frame, the amount of data is: (2 * 6 bit + 2 bit) * 32 = 448 bit ^ 4. same scaling factor for all three sl sl sl sl 11 s2 s2 s2 s3 s3 s3 15 For one frame, the amount of data is: (1 * 6 bit + 2 bit) * 32 = 256 bit

Considering only three time-sequential scaling factors (in the example, the frame corresponds to exactly 24 ms), 20 has the advantage of only a small decoding delay and a small access unit (the smallest unit to be encoded). Unless these advantages are weighted, larger data blocks can be achieved by looking at larger time blocks 25 - by using an input matrix with more columns »- by moving the frame data 'no scf transferred'

Reading example: 30. consecutive scaling factors: 10 40 38 37 36 38 calculated differences: 30 -2 -1 +2 corresponding categories: Is I2 I2 I4 transfer model: scfi scf2 scf2 max (scfi, scf2, scf3) transferred scaling factors: 10 40 38 35 control data: 01 00 decoded scaling factors: 10 40 40 38 38 38

Claims (2)

7 105248 1. A method for transmitting digitized, block-coded audio signals using scaling factors formed from the peak value of a 5-string (block) of signal values in a block-coding of digitized audio signals, associated in quantized form with ) generating the differences di2-n = scfi2 - scfn ^ from the number k of a given frequency subband or group of 10 spectral values of the audio signal over a number of time sequences (scfn, scfi2 / ....., scfik; scfni, scfn2 / ....., scfnk) dik-i (ki) = scfik - scfi (ki) j II · dn2-nl = SCfn2 “SCfnl« · 7 dnk-n (kl) = scfnk “SCfn (kl) taking into account the signs and size; B) separating the (k-1) * n formed in accordance with step a) into at least two value classes, each value category comprising one or more possible differences; c) selecting scaling factors based on a sequence of (k1) * n value classes formed for each n subbands or spectral value groups separately in step b) and providing identification information, wherein the number of consecutive differently selected scaling factors in the queue is less than or equal to the number of different scaling coefficients, and wherein the identifying information identifies each of the 30 selected scaling coefficients in one or more of the signal sample k values of the subband or spectral value group in question; and wherein, on the decoder side, d) assigning the respective selected scaling factors based on the identification information to the blocks of signal sample values, and e) the sample values of the signal and the associated selected ones. the scaling factors are again converted into audio signals which are more or less similar to the original audio signals, characterized in that, within the audio data reduction method that reduces irrelevance and redundancy, the following method steps are arranged in block coding: 5 f) more than (k1) two value categories; g) selecting the scaling factors according to step c), specifying, for each n subbands or spectral group, the transmission model of the new scaling factors based on the psychoacoustic aspects of the pre- and post-attenuation effects of the human auditory perception; where are the new scaling factors located. A method according to claim 1, characterized in that the control information possibly indicates that no new scaling factor is transmitted, but that the previous new scaling factors apply to all of these k blocks. Patentlcrav 20 1. Förfarande för att överföra digitaliserade, blockkodade ljudsignaler under användning av skalfaktorer, som vid block-codningen av de digitaliserade ljudsignalerna bildas av beloppet av toppvärdet för en sequens (block) av signal. och i quantizer form tillfogas de avkända as a signal source i 25 codename sequences, vid i w i t h a n d i n g a) det ur et antal k tidsmässigt efter varandra feljande scalfaktorer (scfn, scf12, ...... scflk; scfni, scfn2, ... ... scfnk) i ett delfrekvensband eller en group spectral color av den in delband eller spectral color uppdelade ljudsignalen Ί-Ό (med n £ 1) image enligt tecken och belopp differenserna di2-n = scf12 - scfn dlk-l (kl) * scflk - scfi (k_i). dn2-n 1 "SCfn2" sc ^ nl ♦ * * / ^ nk-n (kl) * scfnk "SCfn (ki) 35 9 105248 b) de enligt steg a) bildade (kl) 1n differentiation in the color class, omfattar en gaming av en eller flera nightmare differenser. 5 c) bundled image file den (b) image color image scanner, scaled image for the spectral dye group, color selection, image information, color information , olika selection scale scaling factor 10 inom földden om mind än eller lika med antalet efter varandra feljande, olika scalfactor för det actella del-bandet respective spectral group blocken för de 15 avant-garde signal characteristics i det actel delbandet respektive spectral name-group, d) sker det i samband med markörinformationen en samordning av blocken i de avant-garde signal system med tillhörande selecterade scalfactor och 20 e) ur de avordena äter ljudsignalerna, som mer eller all motsvarar de ursprungliga ljudsignalerna, kännetecknat av att vid blockkodningen förekommer feljande förfarandesteg inom det irrelevans- och redundansminskande: 25 1juddatareduktionsförfarandet: f) de enligt steg a) image (a) image (a) image (c) the best scales factor for the var scaling factor; skalfaktorerna, samt 35 h) som markörinformation användes en styrinformation, som anger pä vilka ställen k de nya skalfaktorerna föreligger. 10 105248 2. Förfarande enligt patentkrav 1, kännetecknat av att styr-Informationen eventuellt inneb & r att ndgon ny scalfaktor inte scall örföras, then den föregäende nya skalfaktorn gäller '> f samtliga k blocken. 5 1 v: 'i. ·